For mate-searching species, the learned mate recognition (LMR) hypothesis assumes that sexual harassment favours signal variation among females, which exploits the receiver ability of males. The model predicts that coevolving males have responded to the female sexual foil by learning to recognize female variants as potential mates. I translate the LMR hypothesis into the language of signal detection theory to explain its novelty as a dynamic, coevolutionary, negative frequency-dependent selection model. Due to genee environment interactions, males cueing to the morph detected most often should generate positive but often asymmetrical, detection-dependent harassment towards females. Females are expected to sort to an ideal free distribution where harassment costs are equal. At equilibrium, morph fitness, but not necessarily morph frequency, is predicted to be equal. The LMR hypothesis is consistent with recent experimental data and the distribution of colour polymorphisms in the Odonata, predicts general conditions favouring variation in sexual signals, and provides a novel mechanism for speciation via sexual signalling.